7/31/ D Cone-Beam CT: Developments and Applications. Disclosure. Outline. I have received research funding from NIH and Varian Medical System.

Transcription

1 4D Cone-Beam CT: Developments and Applications Lei Ren, PhD, DABR Department of Radiation Oncology Duke University Medical Center Disclosure I have received research funding from NIH and Varian Medical System. Outline Background Image acquisition approaches Projection sorting techniques Reconstruction techniques Imaging dose Clinical applications of 4D-CBCT 1

2 Background: On-board X-ray based Imaging Techniques Radiograph Cone beam CT (CBCT) Digital Tomosynthesis (DTS) Background - On-board X-ray based Imaging Techniques Q. J. Wu, IJROBP, 2007 Static phantoms Radiograph CBCT DTS (30 ) Scan angle: 0 Scan time: <<1 s Scan dose: <<1 mgy Dimension: 2D Scan angle: 360 /~200 Scan time: ~1 min Scan dose: ~1-8 cgy Dimension: 3D Scan angle: 20 ~ 60 Scan time: < 10 s Scan dose: < 1 cgy Dimension: Quasi-3D Background - Imaging for moving lung 3D CBCT Static 3D CBCT 4D CBCT Moving phantom/ patient 4D imaging needed 2

3 Background: Challenges with 4D-CBCT Image acquisition: Unevenly distributed projections, slow gantry rotation (~1deg/s, 3-4min over 200deg) Projection sorting: robust, automatic, fast Reconstruction: under sampled projection data for each phase Image acquisition approach N Patient 1 θ 2 θ: scan angle for a single phase in one cycle N_cycle: number of cycles scanned Ψ: total scan angle for a single phase G: gantry rotation speed ( 6º/s) T: patient respiratory period N_phase: number of phase bins (typically 10) Ω: total scanning angle (360º for half fan, 200º for full fan) θ = N_phase N_cycle = Ω Ω Ψ = θ N_cycle = N_phase Image acquisition approach N Patient 1 θ 2 θ = N_phase N_cycle = Ω Ω Ψ = θ N_cycle = N_phase G θ N_cycle Ψ no change, projections more spread out G = 6º/s, T = 5s, N_phase = 10, Ω = 360º : θ = 3º, N_cycle =12, Ψ=36º 3

4 Image acquisition approach Patient θ θ = N_phase N_cycle = Ω Ω Ψ = θ N_cycle = N_phase G θ N_cycle Ψ no change, projections more spread out G = 3º/s, T = 5s, N_phase = 10, Ω = 360º : θ = 1.5º, N_cycle =24, Ψ=36º Slow gantry rotation for 4D-CBCT scan to spread out projections to reduce streak artifacts Image acquisition approach T=5s, N_phase = 10, Ω = 360º G=6º/s G=3º/s G=1º/s θ = N_phase Image acquisition approach N_cycle = Ω T θ N_cycle image quality T=6.53s G=2º/s T=4.93s T=3.37s Slower gantry rotation needed for longer respiratory period Li et al, 67(4), , IJROBP, (2007) 4

5 Relative Magnitude Phase (%) /31/2017 Projection sorting techniques 4D-CBCT reconstruction requires sorting of projection images into different respiratory phases. Phase sorting methods: External surrogates: bellow system, RPM system, spirometry Require additional devices External internal mis-correlation Internal surrogates: Marker based and markerless methods Marker based method: invasive implantation, marker migration Markerless methods: Tracking structures with respiratory motion Tracking image intensity fluctuation with breathing Motion modeling method Projection sorting techniques : FT-Magnitude technique Acquire projections Perform 2D FT Extract resp signal Identify peak-insp 800 % % 40% 60% 80% Projection Number Projection Number Assign phase Sort/Bin projections Recon 4D-CBCT Irina Vergalasova et al, AAPM Annual Meeting, 2011 Projection sorting techniques : FT-Magnitude technique Patient #1 x FT Signal FT-Mag Peak-Insp Manual Peak-Insp Projection Number Irina Vergalasova et al, AAPM Annual Meeting,

6 SI w1 % of Phase 7/31/2017 Projection sorting techniques: Motion modeling based method Prior Knowledge On-board volume at time t considered as a deformation of CT Motion model PC1,2,3 9 Deformation fields DVF t 3 PC i w i (t) i=1 No On-board Projection Data Fidelity Constraint Met? Yes w i (t) for projection sorting 4D-CT (10 phases) 3D Prior I prior DRR of deformed CT based on DVF(t) Lei Zhang et al, AAPM Annual Meeting, 2016 Projection sorting techniques: Motion modeling based method 1000 PCA Coefficient SI w Projection Number 100 PCA Coefficient based Phase Sorting Robust against changes in tumor size, location, and breathing pattern changes. This method could be used in both CBCT full fan and half fan mode, and for different gantry rotation speeds. Lei Zhang et al, AAPM Annual Meeting, 2016 Projection sorting techniques: Motion modeling based method Patient Number Average Phase Difference Phase Difference within 10% Phase Difference within 5% P1 (FF) 2.23% 99.4% 88.9% P2 (FF) 1.81% 98.7% 93.4% P3 (HF) 1.62% 100% 99.8% P4 (HF) 1.82% 100% 96.2% P5 (HF) 1.63% 98.4% 96.6% Lei Zhang et al, AAPM Annual Meeting,

7 Reconstruction techniques Feldkamp back projection Iterative method with compressed sensing Prior image based method: MKB, PICCS Motion compensated Prior knowledge with motion modeling Reconstruction techniques: FDK Method Cone Beam---Feldkamp-Davis-Kress (FDK) algorithm A 3D realization of the filtered-back projection algorithm Fast, easy-to-implement Streak artifacts Reconstruction techniques: FDK Method 45 projections 90 projections 180 projections 7

8 Reconstruction techniques: Iterative method with TV ART, SART, SIRT + regularizations (TV) TV f = න f x 1 dx Remove streak artifacts for sparse projection sampling in 4D-CBCT Computationally expensive Image over smoothed ART: algebraic reconstruction technique SART: simultaneous algebraic reconstruction technique SIRT: simultaneous iterative reconstruction technique TV: total variation Reconstruction techniques: Iterative method with TV 45 projections 90 projections 180 projections Reconstruction techniques: McKinnon-Bates (MKB) algorithm Less streaks, no over smoothing Residual motion artifacts Zheng et al, Proc. SPIE. Vol. 7961, (2011). McKinnon et al, IEEE Transactions on Biomedical Engineering, 2, , (1981) 8

9 Reconstruction techniques: McKinnon-Bates (MKB) algorithm Prior FDK MKB MKB+ truncation correction Zheng et al, Proc. SPIE. Vol. 7961, (2011). Reconstruction techniques: prior image constrained compressed sensing (PICCS) Construct prior image Xp by back projecting all phase projections min Ψ 1 X X P l1 + 1 Ψ 2 X l1, s.t. AX=Y FBP TV PICCS Chen et al, Med. Phys., (2008). Reconstruction techniques: Motion compensated 4D- CBCT (MoCo 4D-CBCT) Projections FDK 4D-CBCT 3D CBCT FDK 4D-CBCT MoCo 4D-CBCT Deformation fields MoCo 4D-CBCT Li et al, Phys. Med. Biol., 2006 Brehm et al., RSNA, 2012 Rit et al, IEEE Trans. Med. Img,

10 Reconstruction techniques: Simultaneous Motion Estimation and Image Reconstruction (SMEIR) Wang and Gu, Med. Phys Dang et al, IJROBP, 2015 Reconstruction techniques: Simultaneous Motion Estimation and Image Reconstruction (SMEIR) 39 projections 200 projections Dang et al, IJRBP 2015 Reconstruction techniques: prior knowledge based image estimation technique Prior CT Volume Deformation Field Map (DFM) Estimated onboard images Data fidelity constraint DRRs Match On-board Projections Prior CT Deformation field solved Estimated on-board images Ren et al, Med. Phys.,

11 Reconstruction techniques: prior knowledge based image estimation technique MCBCT new (D, CT) P 2 = 0 projections 1. D ~10 7 variables 2. limited-information acquired (~10 2 projs) ill-conditioned problem need dimension reduction(modeling) or regularization of D Reconstruction techniques: prior knowledge based image estimation technique Data fidelity constraint (PCA) -based motion modeling Coarse estimation of deformation field Fast, accuracy limited by modeling accuracy Data fidelity constraint Free form deformation (FD) with energy constraint Fine-tuned deformation field Slower, accuracy not affected by modeling Reconstruction techniques: prior knowledge based image estimation technique Prior planning CT FDK CBCT (200º) Prior based CBCT (ortho 15º) Prior knowledge based CBCT uses only 1/6 of imaging dose of FDK 4D-CBCT scan Harris et al, Med. Phys.,

12 4D-CBCT imaging dose 4D-CBCT protocol: 200º, 4min, 1320 projections, 20mA PMMA phantom with D=32cm, Length=45cm. TLD every 1.5cm CBDI = 1 Y/2 L න D y dy Y/2 CBDI w = 1 3 CBDI center CBDI peripheral CBDI 100 (mgy) CBDI 450 (mgy) Center 0º 90º 180º 270º CBDI w 12.0± ± ± ± ± ± ± ± ± ± ± ±1.3 Thengumpallil et al, JACMP, 2016 Clinical applications: pretreatment 4D target localization Courtesy of Edwin Sham Clinical applications: pretreatment 4D target localization Dual registration: 1. Register to the Clipbox Courtesy of Edwin Sham 12

13 Clinical applications: pretreatment 4D target localization Dual registration: 2. Register to the mask Courtesy of Edwin Sham Clinical applications: intra-fraction 4D target localization Limited-angle Intrafraction Verification (LIVE) System MV Check point Check point MV Beam #1 Beam #2 Beam #3 kv Dynamic Conformal Arc or VMAT Treatment kv 3D/IMRT Treatment Ren et al, Med. Phys., 2014 Clinical applications: intra-fraction 4D target localization Concurrent kv-mv imaging scheme during treatment 13

14 Summary 4D-CBCT provides on-board 4D images of tumors with respiratory motion to minimize the motion artifacts for target localization. 4D-CBCT requires slow gantry rotation, automatic projection sorting and reconstruction algorithms using under sampled data. Applications for both inter- and intra-fraction 4D verification. Partially supported by NIH: R01CA Varian Medical System Acknowledgements Fangfang Yin, PhD Jing Cai, PhD Devon Godfrey, PhD Wendy Harris, BS Lei Zhang, BS Paul Segars, PhD Jing Wang, PhD Edwin Sham, PhD Xiaobai Sun, PhD Yawei Zhang, PhD You Zhang, PhD Irina Vergalasova, PhD Yingxuan Chen, BS 14